Resting-state fMRI has spurred an impressive amount of methods development, among which dynamic functional connectivity (dFC) is one important branch. However, the relevance of time-varying and time-resolved features has led to debate, to which we want to bring in our viewpoint. We argue that, while statistically many dFC features extracted from resting state are contained within a sufficiently strong null model, these features can still reflect underlying neuronal activity. The use of naturalistic experimental paradigms, at the boundary between resting state and task, is essential to validate their relevance. In parallel, leveraging methods that specifically rely on sparsity is an avenue to strengthen the statistical significance of time-resolved information carried by ongoing brain activity.